Broadband sensor signal processing, detection and classification of rare and unusual events

该研究项目是从宽带传感器和阵列中提取信息，包括声学和电磁。特别感兴趣的是罕见或不寻常事件（异常值）的检测和分类。我们关注三种类型传感器的数据：(1)水下电缆海洋观测站网络中的水听器，(2)由智能手机制成的任意麦克风阵列，(3)宽带无线电天线。我们的方法的一个关键新颖之处在于扩展了我们在水听器数据方面的早期成功，包括其他类型的传感器，并探索了音频信号处理技术在宽带无线电信号中的应用。随着我们从水听器到智能手机再到无线电，背景噪音和不必要的事件变得越来越复杂，风险也随之增加。该研究使用真实世界的测量数据，并以与最终用户的交互为指导。应用范围从声环境监测到海洋生物学，从无线通信到安全。加拿大海洋网络公司在东北太平洋、不列颠哥伦比亚省维多利亚附近的萨利希海和北极地区运营着几个电缆连接的海底观测站。这些观测站目前通过连接到海洋仪器的光纤电缆从海底传送连续的实时信息。水听器的数据是24小时流式传输并存档的。这些数据大多是无趣的杂音。我们面临的挑战是找到稀有（不寻常）的音频事件，如航运和海洋哺乳动物的活动。这项工作利用了加拿大海洋网络公司2.25亿美元的海底电缆观测站投资，以充分利用持续观测提供的机会。在庞大的数据量中，人工地收听、发现和分类罕见和不寻常的事件是不切实际的。提出的罕见事件检测和分类算法对于这些持续的海洋监测项目的成功至关重要。我们考虑的麦克风阵列是任意的，可能会改变几何形状，灵感来自于无处不在的具有GPS定位功能的智能手机。智能手机是一个很有前途的实验平台，因为它们包括扬声器、无线（收音机）和麦克风，而且是可编程的。原则上，这些智能手机可以用来收集声学数据，当人们（用智能手机）聚集在一个地方（室内或室外）时，这可能会特别有趣。许多正在使用的智能手机提供了产生大量数据的潜力。智能手机阵列的功能包括估计声源位置和/或房间特性（尺寸、材料反射系数）、去混响、声源分离、语音增强。这些功能可用于支持和增强罕见事件的检测和分类，从而利用无处不在的智能手机在音频环境监测方面创造新的功能。超宽带（UWB）无线电系统已经得到了广泛的研究，并已商业化。美国规定了最大频谱密度（单位带宽功率）。大多数超宽带设备在3- 10ghz范围内工作，但在1ghz以下允许使用相同的频谱密度。在这个较低范围内的超宽带系统很少被研究。这个无线电频谱也有我们希望检测和分类的稀有事件（例如，未经授权的传输，不寻常的传播现象）。这些超宽带无线电系统可以被视为类似于音频系统，因为它们的带宽跨越超过10个八度，因此用于水听器和智能手机的罕见事件检测技术可以适用于无线电。应用包括无线通信、基础设施监控和监视。

The research program is the area of extracting information from broadband sensors and arrays, both acoustic and electromagnetic. Of particular interest is the detection and classification of rare or unusual events (outliers). We focus on data from three types of sensors: (1) hydrophones in underwater cabled ocean observatory networks, (2) arbitrary microphone arrays made with smartphones, and (3) broadband radio antennas. A key novelty of our approach is to expand upon on our early success with hydrophone data to include other sensor types, and exploring the application of audio signal processing techniques to broadband radio signals. The background noise and unwanted events become increasingly complex as we move from hydrophones to smartphones to radio, with associated increase in risk. The research uses real-world measured data and is guided by interaction with end users. Applications range from sound environment monitoring to marine biology to wireless communications to security. Ocean Networks Canada operates several cable-linked seafloor observatories in the Northeast Pacific Ocean, Salish Sea near Victoria BC and the Arctic. These observatories currently deliver continuous real time information from the seafloor via fibre optic cables connected to oceanographic instruments. Hydrophone data is streamed 24/7 and archived. Most of this data is uninteresting noise. The challenge is to find the rare (unusual) audio events of interest such as shipping and marine mammal activity. The work leverages the $225M investment in Ocean Networks Canada cabled undersea observatories, to fully exploiting the opportunity provided by continuous observation. It is impractical to manually listen, find and classify rare and unusual events in the huge volume of data. The proposed rare event detection and classification algorithms are essential for the success of these sustained ocean monitoring programs. We consider microphone arrays of arbitrary and possibly changing geometry, inspired by the ubiquitous use of smart cellphones with GPS location abilities. Smartphones are a promising platform for experimentation since they include speakers and wireless (radios) as well as microphones and are programmable. In principle, these smart cellphones may be used to gather acoustic data, which may be particularly interesting where people (with smartphones) are gathered in one place (inside or outside). The many smartphones in use offer the potential to generate a huge volume of data. Smartphone array capabilities range from estimating source locations and/or room characteristics (dimensions, material reflection coefficients), de-reverberation, sound source separation, speech enhancement. These capabilities can be used to support and enhance the detection and classification of rare events, thus using the omnipresence of smartphones to create new capabilities in audio sound environment monitoring. Ultra-wideband (UWB) radio systems have been studied extensively and are commercially available. The regulations in the US specify the maximum spectral density (power per unit bandwidth). Most UWB equipment operates in the 3-10 GHz range but the same spectral density is permitted below about 1 GHz. UWB systems in this lower range are much less commonly studied. This radio spectrum also has rare events of interest (e.g. unauthorized transmissions, unusual propagation phenomena) that we wish to detect and classify. These UWB radio systems may be viewed as analogous to audio systems in that their bandwidth spans more than 10 octaves, thus the techniques for rare event detection used with hydrophones and smartphones may be adapted for radio. Applications include wireless communications, infrastructure monitoring and surveillance.